For many numerical modelling applications the problem of specifying an opti
mum mesh resolution remains unbounded and for mesh construction objective a
priori rules do not exist. By contrast, the problem of specifying model pa
rameter surfaces is largely bounded within known physical error distributio
ns. In this paper we thus investigate the impact of varying mesh resolution
on a typical non-linear finite numerical solver. Specifically, a two-dimen
sional finite element code which solves the Shallow Water equations was use
d to simulate unsteady flows in a meandering compound channel. A range of d
ifferent mesh resolutions and parameter surfaces were simulated to determin
e relative dominance and, unlike previous studies, the effect on both bulk
flow and distributed outputs were analysed. The results showed a wide varia
tion in performance for mesh discretizations which fulfilled traditional le
ngth scale-based construction. Mesh resolution effects were at least as imp
ortant as a typical calibration parameter and model response was shown to b
e highly complex. (C) 1999 Elsevier Science B.V. All rights reserved.